Prediction Market Research, Newsletters, and Data Sources

Prediction-market research needs three independent layers: contract truth, market data, and analysis that preserves what was knowable at the time.

A probability-looking price is seductively compact. It can also hide the contract wording, fee schedule, thin order book, access rules, resolution source, and moment at which the evidence became public. A useful research stack makes those details easier to inspect instead of turning every price move into a story about collective wisdom.

I reviewed the public sources named here on September 4, 2026. Inclusion requires an official or publicly inspectable source, a distinct role in the workflow, and enough documentation to identify limits. This is not a ranking by returns. Alphanume Research is included with disclosed affiliation, and Alphanume does not claim to provide a complete prediction-market data feed. Platform availability, contracts, fees, and legal eligibility can change, so verify current first-party terms before using any venue.

Start with the contract, not the forecast

The CFTC's prediction markets explainer says event contracts are commonly based on future outcomes and often use yes-or-no structures with a fixed payout. It also tells customers to expect clear information about terms, settlement, risks, fees, and who makes the settlement determination. That is the correct research order.

Save the exact title, complete rules, listed resolution source, close condition, settlement procedure, and every rule revision. Two contracts with similar headlines may resolve differently. A contract asking whether an agency reports a value by a date is not equivalent to one asking whether the underlying event economically occurred.

• Resolution source: Which named authority or publication decides the outcome?
• Observation rule: Is the value initial, revised, seasonally adjusted, certified, or final?
• Time zone: Which clock governs trading close and the qualifying event?
• Ambiguity process: Who interprets language when facts do not map cleanly to the contract?
• Revision history: Can you recover the rules that traders saw before later clarification?

Use official venue data for the market layer

Kalshi's official API documentation describes an order-book endpoint that returns active bid orders for both sides of a binary market. That makes it a direct source for current book state when the contract and your access are eligible. Check its current authentication, historical-data, rate-limit, and fee documentation before designing a production collector.

Polymarket's official market-data overview documents the public market-data and central-limit-order-book interfaces for its Polygon-based product. That product should not be conflated with Polymarket US. As of September 4, 2026, the CFTC trading-organization registry lists QCX LLC doing business as Polymarket US as a designated contract market. Venue architecture, eligibility, products, and terms must be checked for the specific entity and interface under study.

Newsletters are an interpretation layer

The best research letter explains information arrival, participant incentives, market design, and failure cases. PredictionMarkets.org describes a free weekly newsletter covering platform changes, regulation, and market signals. The Oracle by Polymarket is a venue-owned newsletter built from Polymarket markets, so its direct market access is a strength and its platform affiliation should remain visible. Gemini Predicted is another venue-affiliated publication using prediction-market data for news analysis.

These were reviewed through official public pages on September 4, 2026. Alphanume has no stated affiliation with them. They are listed by research role rather than claimed returns: use the independent weekly brief for sector discovery, and the venue publications for examples of how operators interpret their own market data. None should replace the venue rulebook or the named resolution source.

A good issue leaves you with a testable question. Did an official update move the best executable price, or only the last trade? Was the market slow relative to a primary release after fees? Did two related contracts temporarily violate a logical constraint? Those questions can be timestamped and falsified. "The crowd was wrong" usually cannot, unless the author defines when, at what price, and relative to which information set.

When evaluating a newsletter, sample old issues and compare publication time with the venue record. Check whether edits are visible and whether losing or ambiguous cases remain in the archive. A strong explanatory article may be valuable even when it has no trade, as long as it does not dress hindsight up as a live forecast.

Preserve a research-ready event log

Prediction markets combine market microstructure with real-world data engineering. You need the book and trades, yet you also need a timestamped external evidence feed. The difficult field is often not price. It is when a qualifying fact became publicly observable under the exact contract rule.

prediction_market_observation

venue
market_id
contract_title
rules_version_hash
rules_captured_at_utc
resolution_source_url
market_status
best_yes_bid_size
best_yes_ask_size
last_trade_price
fee_schedule_version
external_fact_published_at_utc
external_fact_collected_at_utc
resolved_outcome
settled_at_utc
data_gaps[]

Store both published and collected timestamps for external facts. If an agency silently revises a page, a backtest built from the current version may know more than a trader did. Hashing the rules and source payload gives you evidence that a later semantic change did not leak backward.

Turn probability claims into executable tests

A quoted yes price is not automatically a calibrated probability. Fees, spread, order size, collateral treatment, trader preferences, and access constraints can separate price from a frictionless belief. For a calibration study, choose a consistent price convention and report results by liquidity and horizon.

1. Freeze the eligible contract universe before examining outcomes.
2. Use a timestamped bid, ask, or midpoint rule and state why it matches the question.
3. Separate observations from the same event family to avoid pretending they are independent.
4. Account for fees, settlement timing, invalid markets, and capital tied up until resolution.
5. Report calibration bins, count, spread, depth, and missing-history coverage.
6. Test whether apparent edge survives after the public source timestamp and realistic latency.

Logical-arbitrage research needs equal care. Mutually exclusive outcomes may sum to an attractive number at displayed quotes while the necessary size is unavailable, one leg closes earlier, or rules leave a state outside the set. Execute the semantics on paper before executing the orders.

Where Alphanume Research fits

Alphanume Research contributes worked investigations rather than a complete venue archive. We Figured Out How Prediction Market Quants Actually Trade examines behavioral and data clues. We Found Another Edge in Prediction Markets turns a market observation into a more systematic question. Read both as research examples whose assumptions still need verification against current rules and data.

Bonding Bots in Prediction Markets is useful for seeing how automated behavior and market design interact. It is not a claim that a historical pattern will persist. The durable lesson is to preserve book state, identify the actor hypothesis, and check the behavior after costs and rule changes.

Our dataset catalog, documentation, and validation page show how we define and audit Alphanume products. They do not advertise a comprehensive prediction-market feed. Use official venue APIs for that market layer and use Alphanume's public articles as hypotheses or examples.

Build a stack that can survive a dispute

The minimum stack is a regulator page for status and obligations, official venue rules and data, the named resolution source, and your own immutable observation log. Add newsletters only when they improve interpretation or method. Add third-party aggregators only after checking their identifiers, timestamps, rule versions, and missing-market policy.

Before trusting a result, replay one disputed or ambiguous market from raw records. Can you recover what the contract said, what the book showed, what the official source published, and why settlement followed? If not, your historical accuracy statistic rests on a database that cannot explain its hardest row.

Prediction-market research becomes high quality when it is less impressed by the probability and more obsessed with the contract. Preserve the words, clocks, quotes, sources, and revisions. Then let the analysis earn its confidence one timestamp at a time.